SpecQ Sales Intelligence Package — Electronic Chemicals Sales Skill v1.1
Version: v1.2 | Updated: 2026-06-24 | Added cross-platform installation guide (OpenClaw / WorkBuddy)
📦 Installation Guide
OpenClaw
This Skill has been integrated into OpenClaw MCP:
openclaw skills install daizehua-wq/specq-intel-sales
MCP Server Address: http://119.91.223.127:8001/mcp (HTTP SSE, X-API-Key configured)
WorkBuddy (Tencent Desktop Agent)
Search for specq-intel-sales in the left "Skills" panel of WorkBuddy and install this Skill.
After installation, copy and paste the following text to WorkBuddy, and it will automatically configure the MCP connection for you:
Help me configure the SpecQ MCP server: download specq_mcp_client.py from https://github.com/daizehua-wq/Specq-mcp to the local machine, then create ~/.workbuddy/mcp.json and add the specq configuration, command python specq_mcp_client.py, environment variable SPECQ_MCP_URL=http://119.91.223.127:8001/mcp, SPECQ_MCP_API_KEY=fill in your key. Then pip install mcp httpx.
Use Cases
Before a meeting with a client, electronic chemical sales engineers / pre-sales engineers can quickly obtain a structured sales intelligence package.
Trigger Conditions
The skill triggers when the user provides any of the following sets of information:
| Trigger Mode | Example |
|---|---|
| Three Fields | "Help me create an intelligence package for XX electroplating solution, customer Shennan Circuits, they care about roughness" |
| History Query | "What was discussed with Shennan Circuits before?" |
| Feedback | "Shennan Circuits closed the deal" / "Lost the deal, they thought the price was too high" |
| Multi-task | "Help me create three intelligence packages: Shennan + Copper Plating, Jingwang + Etching Solution, Pengding + Electroless Copper" |
Dependencies on MCP Server (5 Tools)
| Tool | Input | Output | Description |
|---|---|---|---|
specq_memory | action, query/content/category... | Memory operation result | v1.1 New: Three-layer memory |
specq_generate_intel | product, application, scenario | Eight-module intelligence package Markdown | Core generation |
specq_extract_insights | customer_id (optional), limit | Structured insights JSON | Dark data extraction |
specq_log_visit | customer_id, content, visit_date, visit_type | Visit record ID | Log visits |
specq_feedback | product, application, outcome, lesson, accuracy_notes | Feedback record | Win/loss closure |
v1.1 Workflow (Including Three-layer Memory)
User Input
↓
① Working Memory Check (specq_memory.get_plan)
├─ Has unfinished task → Prompt: "Last XX wasn't completed, continue?"
└─ New task → Continue
↓
② Long-term Memory Recall (specq_memory.recall)
→ Semantic search for historical visits, lost deal records, preferences for this customer/product
↓
③ Extract Three Fields (product / application / scenario)
→ if three fields are incomplete → Follow up with a question
↓
④ Dark Data Injection (extract_insights + memory recall results)
├─ Have dark data → Inject into intelligence package
└─ No dark data → Mark as [Experience-based Inference]
↓
⑤ Generate Intelligence Package (specq_generate_intel)
↓
⑥ Output Eight-module Intelligence Package
【Historical Memory】(Independent context block)
- 2026-06-20 Visit: Focus on roughness, competitor Atotech
- 2026-05-15 Lost deal: Price 20% too high
【Current Intelligence Package】(Eight modules)
... (With data source annotations)...
↓
⑦ Write to Long-term Memory (specq_memory.save)
→ category=intel, content=Summary of current intelligence package
↓
⑧ Update Working Memory (specq_memory.set_plan)
→ Mark current task as complete
↓
⑨ ⚠️ Reminder: "Let me know the outcome after you meet the client, so I can record it."
Scenarios
Scenario A: New Customer (No Historical Data)
- Long-term memory recall returns empty → Call generate_intel directly
- Modules 4/5/6 are annotated with
[Experience-based inference, no sales data to support]
Scenario B: Returning Customer (Has Memory Data)
- First recall → Retrieve historical visits + lost deal records
- Inject into the independent context block【Historical Memory】
- Intelligence package modules 4/5/6 have actual data source annotations
Scenario C: Pure History Query
- recall → Display historical memory in a structured format
- After display, follow up with: "Generate an intelligence package based on this information?"
Scenario D: Win/Loss Feedback
- feedback → Record the closure
- Simultaneously save(category="feedback") to long-term memory
- outcome: won / lost / follow_up
Scenario E: Multi-task (New in v1.1)
- get_plan checks progress → Knows what has been done
- set_plan sets/updates the task list
- Resume after interruption: User says "continue" → get_plan → Knows where to start
Memory Category System
| category | Trigger Event | Example Content | Retention Policy |
|---|---|---|---|
visit | Automatically written after log_visit | "Visit XX customer, focus on YY metrics" | Permanent |
feedback | Automatically written after feedback | "Lost deal for XX product, reason: 20% higher price" | Permanent |
intel | Automatically written after generate_intel | "Generated YY intelligence package for XX customer" | 180 days |
insight | Structured discovery from extract_insights | "Shennan Circuits' sensitivity to roughness: High" | 180 days |
preference | User explicit preference | "David prefers to see competitive comparisons first" | Permanent |
Output Specification
Intelligence Package (Eight Modules)
| Module | Content | Data Source |
|---|---|---|
| 1. Product Overview | Product definition, core functionality, applicable process steps | Public data + knowledge.db |
| 2. Technical Benchmarking | Key parameters vs. competitors/industry standards | Public data + dark data |
| 3. Competitive Landscape | Key competitors, market share, differentiation | Public data |
| 4. Client Focus Metrics | Key technical metrics for this client/industry | Dark data / [Experience-based inference] |
| 5. Entry Opportunities | Current entry window for this client | Dark data / [Experience-based inference] |
| 6. Adoption Barriers | Historical reasons for lost deals, technical hurdles | Dark data / [Experience-based inference] |
| 7. Action Recommendations | Visit talking points, demo focus, pricing strategy | LLM synthesized |
| 8. Reference Sources | Data sources for each module + confidence level | Auto-annotated |
Historical Memory (Independent Context Block)
Every time an intelligence package is generated, attach a historical memory block at the very beginning of the output:
【Historical Memory】
- 2026-06-20 Visit Shennan Circuits: Concerned about roughness (Ra<0.3μm), competitor Atotech
- 2026-05-15 Lost deal: Price 20% too high, suggest price concession in next quote
Important Rules
- Memory Priority: Always call recall at the start of every conversation. Never skip if memory exists.
- Source Transparency: Annotate data sources for each module. Annotate source for dark data; annotate inferences as
[Experience-based inference]. - Feedback Reminder: Always remind about feedback after outputting an intelligence package.
- Incomplete Parameters Follow-up: When the three fields are incomplete, ask for clarification; do not guess.
- Confidentiality: Intelligence packages do not contain real customer names; anonymize them to industry labels.
- Working Memory Retention: Record progress for multi-tasks; when the user says "continue," the state should be recovered.
MVP Boundaries
Do:
- Cover Scenarios A/B/C/D/E completely
- Automatic read/write of three-layer memory
- Pure Markdown output
Do Not:
- Multi-turn deep-ask for requirements (single-turn priority)
- Automatically push intelligence packages to Feishu/Email
- Real-time competitor pricing
- Memory merging/deduplication